<table border="1" cellspacing="0" cellpadding="8">
    <tr>
        <th>Issue</th>
        <td>
            <a href=https://github.com/llvm/llvm-project/issues/59640>59640</a>
        </td>
    </tr>

    <tr>
        <th>Summary</th>
        <td>
            Regression: stdatomic.h should not forward to MSVC's incompatible version in C11
        </td>
    </tr>

    <tr>
      <th>Labels</th>
      <td>
      </td>
    </tr>

    <tr>
      <th>Assignees</th>
      <td>
      </td>
    </tr>

    <tr>
      <th>Reporter</th>
      <td>
          davidben
      </td>
    </tr>
</table>

<pre>
    This is a regression from e0c3142af075e2ef89395dbed5939071345eb622. (@compnerd FYI)

Originally, Clang's `<stdatomic.h>` would forward to the system's copy if `__has_include_next(<stdatomic.h>)`. This meant that, among other things, clang-cl would still provide C11 atomics even though MSVC headers didn't provide it. (This is nice and consistent with clang-cl defining `__STDC_VERSION__ >= 201112L` without `__STDC_NO_ATOMICS__`.)

Then C++23 decided to add `<stdatomic.h>` to improve C11 compatibility. Newer MSVC now provides stdatomic.h, but it only works in C++23. In C, it fails with:

> #error <stdatomic.h> is not yet supported when compiling as C, but this is planned for a future release.

Ironically, this C11-compatibility improvement in C++23 made clang-cl less C11-compatible. 😄 So, 1ad7de9e92bc2977698e5f6d6493202b50c912d5 made clang's `<stdatomic.h>` skip the system's  if `defined(_MSC_VER)`.

Then ba49d39b20cc5358da28af2ac82bd336028780bc defined the workaround to only skip on MSVC and C. MSVC and C++ would still forward to MSVC's C++23 header. That's still fine and C11-compatible.

However, e0c3142af075e2ef89395dbed5939071345eb622 regressed this. It now targets the workaround to only MSVC and pre-C++23. It seems to have inadvertently dropped the workaround in C11, despite the comment still referencing C. This now breaks clang-cl in C11 mode. clang-cl still defines `__STDC_VERSION__ >= 201112L` without `__STDC_NO_ATOMICS__`, yet `#include <stdatomic.h>` doesn't work.

I think the condition should have been `!(defined(_MSC_VER) && (!defined(__cplusplus) || __cplusplus < 202002L)`
</pre>
<img width="1px" height="1px" alt="" src="http://email.email.llvm.org/o/eJysVk1v2zgQ_TX0ZRBBIq2vgw-pUmMDtA3QBAX2ZFDi2OKGIgWSstf_fkHKrpWkXexhD0kcYTh87817I3Pn5EEjbkj-ieQPKz753tiN4EcpWtSr1ojz5qWXDqQDDhYPFp2TRsPemgEw7Vi2pnyfljlS3Fc1q3PRoshrVqdlxtY5tgWlCRBakXXamWHUaAVs_3wktCbpA0nv599PVh6k5kqdCW2gUVwfCC0dkCIlrHFecG8G2SU9YZ9JkcLJTErA3tgTtwK8Ad8juLPzOMRznRnPIPfh_G7Xc7eTulOTwJ3Gv31A86EprUmRJhDZDsi1B99zH9DwwegDGN-jBd9LfXDhaRcw3nXqAsV5qRSM1hylQGiyDObmDvCIGnxvpkMPX59_NNAjF2gdCCk0oaX_eUr6qNRVcC07BK4FdEY76TxqDyfp-9vVAvdSS32YaT6_PDS7H5-_Pz8-fdvtIJBiD0DTLMvolyiaDDD8ovrb0-7-5enrY_O82wX676by0qOGhtBPhH6iDAR2UmCUmwvx29l4A3IInGYdwtS5l61U0p8T-IYntLMQ2pyu3B0s-9AG2smD9GC0OsPJ2FcHcgElgcf4XxNq9lwqF8kRdr-ET9hnIJShtcbCB6hRYuPhjB7cNI7GehRwCpQDZKmCsNzN1wQ4_jKXUXGtMboPOOwnP1kEiwq5w2R5_6M1WnZXU8fjTZbdvRHkKtUQprukCAMXeBu1Qvf2tMIEyDYl9ZbUFanW8GzCLRkXpcAaa9p2tC7Loq4w3xeiWNeMprTN067OqMgX7f81aO5Vju_DdQlWdB8KQqvd1-fovEuIPlio5etasLqladflLK8EpxXfU95VtBWMFSmtyiptO7i0jBeGqXNrJh0NF40QwRg9uycko0kWn2fl3uRxsSBCXUR_U3gOYsh8CHrproeknnP3Tu4lrT_MCY9og-L_dQte12ekJ10Cjz4mwHN7QO9-x_knv9Hi3TIAHhzi4EJhz48IUnNxRBvWhDqDsGYcPyoZHJZlAbZAN0qPsaAzQ_TfzN_iHi3qLvi_uWzEALS1yF_dzZJzLxiMwOT2dO4xD9L9X4spAA45jR_ZZZd_jHToJQy6ea0G2m_zGNf364WyFtKHV5nro2Gihi2ini_JCK1-aXAgtCC0iK80mi1Kdt2oJhd-YlXZkLKBxdMAF2hK05R-mYOyEhsmalbzFW6yoswKVtAsX_WbEmsm9jVmaZauBbKS79OizSqGfF2xtl7JDU0pzSjNspwyliesK7moad2KLmtFIcg6xYFLlSh1HBJjDyvp3ISbvC7W6UrxFpW7vvXtJhTdtdPBkXWqpPPudsxLr3Dz_eebn7D75aq-qhcW6a_CJvUtQHBEG788zMZZTVZteu9HF9Y23RK6PUjfT23SmYHQbYBw-XM3WvMXdp7QbaThCN1GJv8EAAD__6vkxuQ">